Heteromeric, but Not Homomeric, Connexin Channels Are Selectively Permeable to Inositol Phosphates*

Previous work has shown that channels formed by both connexin (Cx)26 and Cx32 (heteromeric Cx26/Cx32 hemichannels) are selectively permeable to cAMP and cGMP. To further investigate differential connexin channel permeability among second messengers, and the influence of connexin channel composition on the selectivity, the permeability of inositol phosphates with one to four phosphate groups through homomeric Cx26, homomeric Cx32, and heteromeric Cx26/Cx32 channels was examined. Connexin channels were purified from transfected HeLa cells and from rat, mouse, and guinea pig livers, resulting in channels with a broad range of Cx26/Cx32 aggregate ratios. Permeability to inositol phosphates was assessed by flux through reconstituted channels. Surprisingly, myoinositol and all inositol phosphates tested were permeable through homomeric Cx32 and homomeric Cx26 channels. Even more surprising, heteromeric Cx26/Cx32 channels showed striking differences in permeability among inositol phosphates with three or four phosphate groups and among isomers of inositol triphosphate. Thus, heteromeric channels are selectively permeable among inositol phosphates, whereas the corresponding homomeric channels are not. There was no discernible difference in the permeability of channels with similar Cx26/Cx32 ratios purified from native and heterologous sources. The molecular selectivity of heteromeric channels among three inositol triphosphates could not be accounted for by simple connexin isoform stoichiometry distributions and therefore may depend on specific isoform radial arrangements within the hexameric channels. Dynamic regulation of channel composition in vivo may effectively and efficiently modulate intercellular signaling by inositol phosphates.

Connexin channels, which compose most vertebrate gap junctions, mediate direct intercellular movement of ions and molecules. There are ϳ20 isoforms of connexin protein, each forming channels with distinct functional properties. Every known functional deletion of a connexin isoform produces a distinct pathology, and genetic replacement of one connexin (Cx) 2 by another ("knock in") fails to fully compensate (1)(2)(3). Pathologies that arise from altered connexin channel function must arise from abnormal molecular movement through connexin channels, whether in magnitude, regulation, or molecular specificity (reviewed in Ref. 4).
Gap junction channels form by end-to-end interaction of hemichannels, each consisting of six connexin monomers. Hemichannels are either homomeric (composed of a single connexin isoform) or heteromeric (more than one isoform). Dramatic and surprising degrees of ionic and molecular permselectivity have been observed for homomeric channels (5)(6)(7)(8)(9)(10)(11)(12)(13). However, most cells express more than one connexin, and heteromeric connexin channels are common in vivo (14 -18). Heteromeric mixing of different connexin isoforms, producing variation in channel stoichiometry and/or arrangements of isoforms within the hemichannels, may allow cells to dynamically regulate their intercellular communication properties, including molecular selectivity. There is experimental evidence that this is the case (19 -25).
We have reported previously the molecular selectivity between cAMP and cGMP by heteromeric Cx26/Cx32 channels (20,21). Specifically, we found that in populations of native heteromeric channels, composed of channels with a range of isoform stoichiometries and arrangements, some of the channels were permeable to cAMP, and a much larger fraction of the channels was permeable to cGMP. This selectivity between highly similar biological signaling molecules was unanticipated from dye permeability studies of these channels. The ability of connexin channels to discriminate among second messengers is likely to be a biologically important property. Therefore, we asked whether the selective permeability among second messengers extended to inositol phosphates (IPs).
IP permeability is difficult to assess in cells because of the biological effects of manipulation of the levels of these compounds and their short lifetimes in cytoplasm. For example, the lifetime of inositol 1,4,5triphosphate ((1,4,5)-IP 3 ) in cytoplasm ranges from 9 to 60 s, depending on cell type (26,27). For this reason, we utilized a well characterized liposome-based technique for assessing permeability of functional connexin channels (20, 28 -34). Permeability to each IP was determined by direct assessment of its loss and/or retention in liposomes containing functional connexin channels. In this way, the identity of the permeable molecular species was unambiguously determined.
The use of IPs as permeability tracers has several advantages. First, they are endogenous cytoplasmic molecules available for permeation through gap junction channels in vivo. Second, several IPs, in addition to (1,4,5)-IP 3 , have been shown to be biologically active (35)(36)(37)(38)(39)(40)(41). Third, the availability of IPs with different numbers of phosphate groups at several positions around the inositol ring allows exploration of the structural correlates of any selective permeability observed.
To reveal the effect of connexin channel composition on the selective permeability, two strategies were employed to obtain homomeric Cx26 and Cx32 channel populations, and heteromeric channel populations with a wide range of aggregate ratios of the two connexins, from both native and heterologous sources. We report here the surprising ability of heteromeric, but not homomeric, connexin channels to distinguish among biologically active IPs, including those with the same molecular weight and charge. The selectivity is not a simple function of the content of one or the other of the connexins but is a function of the heteromeric character of the channels. This has clear implications for the biological role of heteromeric channels and for modulation of intercellular molecular signaling.
Expression of Tagged Connexin Channels in HeLa Cells-Bi-directional tetracycline-responsive expression vectors (Clontech), containing one or two multiple cloning sites, were used to express homomeric or heteromeric connexin channels in HeLa cells, which have virtually no endogenous connexin expression (43). For expression of homomeric channels, rat (r) Cx26 or Cx32 coding sequences were subcloned into one cloning site in-frame with a sequence coding for a carboxyl-terminal (CT) tag consisting of a thrombin cleavage site followed by a hemagglutinin epitope (not His-Ala) and (His-Asn) 6 , i.e. a HA(HN) 6 tag. When both cloning sites contained different connexin coding sequences, only one connexin was tagged. "Tet-On" cell lines were maintained in 200 g/ml hygromycin and 100 g/ml G418. Channels are designated as Cx26Tag or Cx32Tag when homomeric and designated as Cx26Tag/ Cx32 or Cx26/Cx32Tag when heteromeric.
The supernatant (100,000 ϫ g av for 30 min at 4°C) was incubated with 0.25 ml of agarose-immobilized anti-HA mouse IgG overnight at 4°C with shaking. The antibody matrix was collected at 700 ϫ g av for 1 min at 4°C and washed in a fritted column with 20 ml of 10 mM phosphate-buffered saline, 1 M NaCl, 80 mM OG, 1 mg/ml azolectin, pH 7.4, followed by 20 ml of the same solution containing only 138 mM NaCl. Hemichannels were eluted with 50 mM CH 3 COOH⅐Na, 0.5 M NaCl, 10 mM KCl, 1 mM EDTA, 80 mM OG, pH 4.0, and 0.6-ml fractions were collected into 0.05 ml of 1 M NaHCO 3 , 10 mM KCl, 80 mM OG, pH 9.0. The final pH was ϳ7.4.
Tag Cleavage-Purified protein (200 l) was incubated with 2 units of restriction grade thrombin (Novagen Inc., Madison, WI) for 18 h at 4°C (30). Tag cleavage leaves four extra amino acids at the CT (Leu-Val-Pro-Arg) part of the cleavage site. For these studies, there was no statistical difference in permeability of any of the compounds tested between tagged and tag-cleaved channels ( p Ͻ 0.05; n ϭ 5).
From solubilized mouse and guinea pig membranes, the channels purified are heteromeric Cx26/Cx32. The proportion of Cx26 recovered is greater from guinea pig than from mouse livers, consistent with the composition in the livers themselves (46 -49).
Obtaining Different Aggregate Cx26/Cx32 Ratios from Native Tissue-To obtain different ratios from a single native source, we took advantage of the fact that purification was on the basis of the presence of Cx32. After the connexin was bound to the anti-Cx32 immunobeads, we found that the earliest elution fractions contained the highest proportion of Cx26, presumably due to the lower number of Cx32 epitopes per channel. For example, from solubilized rat liver membranes, the bulk of connexin channels purified are homomeric rCx32. However, when small (ϳ200 l) elution fractions are used instead of the usual ϳ600-l fractions, the first several fractions contained heteromeric rCx26/Cx32 channels with significant Cx26 content, which elute more easily from the antibody matrix. The later fractions contain only homomeric rCx32 channels. This allows a range of isoform stoichiometries to be obtained from a single native source.
Transport-specific Fractionation Activity Assay-Transport-specific fractionation (TSF) was used to assess the molecular selectivity of reconstituted connexin hemichannels (21,28,34,50). TSF fractionates liposomes containing reconstituted connexin channels into two populations within an iso-osmolar density gradient, based on channel functionality. Fractionation is based on channel permeability to urea and sucrose, uncharged solutes that permeate open connexin channels and have different density at iso-osmolar concentrations (459 mM urea or 400 mM sucrose; 500 mosM/kg). Equilibration of these solutes across the liposome through an open hemichannel occurs rapidly and increases the density of the liposome. TSF therefore reports all-or-nothing per-meation of urea and sucrose through connexin channels on a per liposome basis. Other entrapped molecules that are permeable through the channels will be lost from the liposomes during the TSF spin. This means that when the tracer/liposome ratio for liposomes with channels (TSF lower band) is less that that for liposomes without channels (TSF upper band), the difference reflects the number of liposomes in the former population that have channels permeable to the tracer. Fractionation of liposomes into bands is monitored by the fluorescence of rhodamine-PE ( ex 570 nm and em 590 nm) in the liposome membrane.
Typically, TSF uses linear iso-osmotic gradients and centrifugation at 37°C (20, 29 -34). However, for these studies, step iso-osmolar density gradients were formed at 4°C in Beckman SW60Ti 4.4 ml ultracentrifuge tubes (Fig. 1). The buffer solutions were the urea buffer mentioned above and "sucrose buffer," in which the urea was osmotically replaced with sucrose. The step gradients were composed, from the bottom up, of 1.4 ml of 10:90 urea/sucrose mixed buffers, 2 ml of 55:45 urea/sucrose mixed buffers, and 0.4 ml of 90:10 urea/sucrose mixed buffers (all v/v). An aliquot, typically 400 l, of connexin proteoliposomes loaded with tracer molecules by gel filtration (see below) was layered onto each gradient. Gradients were centrifuged at 300,000 ϫ g av for 3 h at 4°C. Liposomes impermeable to sucrose are buoyed by the lighter entrapped urea, banding at the interface between the upper two layers. Liposomes containing functional channels continuously equilibrate the internal and external solution, moving to a position in the gradient corresponding to the density of the liposome phospholipid membrane. These liposomes band at the interface between the middle and the bottom layer. Tracer permeability was assessed by direct comparison of tracers selectively lost or retained by the liposomes in the upper versus lower TSF bands. Tracer retention in the upper liposomes acts as an internal control for tracer loading and for nonspecific membrane leakage.
Tracer Recovery-Upper and lower liposome bands in ϳ100-l volumes recovered from TSF density gradients were analyzed for myoinositol or inositol phosphate content. Liposomes were diluted 1:1 v/v with high pressure liquid chromatography-grade methanol and vortexed. The mixture was slowly passed through a Sep-Pak tC 18 solid phase extraction cartridge (Waters Associates), pre-equilibrated in methanol, followed by 50% v/v methanol in water to elute the tracer. These samples were dried under argon. As an index of lipid removal, fluorescent rhodamine-PE lipid was completely removed from tracer samples by this technique. Residual inositol or inositol phosphate did not bind to the tC 18 matrix solid phase extraction cartridge, as quantitative recovery of unentrapped tracer mixed with liposomes versus untreated tracer samples was obtained ( p 0.05; n ϭ 3).
Tracer Analysis, Enzymatic-Inositol phosphates were dephosphorylated to myoinositol, which was detected using the method of Maslanski et al. (51). Briefly, fresh alkaline phosphatase (45 l of 200 units/ml in 0.1 M Tris-HCl, pH 9.0, 0.1 mM ZnCl 2 ) was added to the dry sample and incubated overnight at room temperature. The enzyme was inactivated at 100°C for 4 min, and tubes were cooled to 37°C. To remove possible contaminating traces of glucose, hexokinase (10 l of 200 units/ml in 50 mM Tris, 10 mM MgCl 2 , 0.1 M ATP, 0.02% w/v BSA, pH 9.0) was added to each sample and incubated at 37°C (1 h). The hexokinase reaction was stopped by 100°C for 3 min. Tubes were cooled to room temperature; NAD ϩ (10 l of 0.1 M) and inositol dehydrogenase (10 l of 1 unit/ml in 10 mM phosphate, 0.02% w/v BSA, pH 6.8) was then added, and incubation was continued at room temperature for 15 min. The pH was then lowered to ϳ6.5 by addition of 5 l of 0.80 M HCl. Resazurin (10 l of 20 M, in 1.0 M phosphate, pH 6.5) and diaphorase (5 l of 10 units/ml in 20 mM phosphate, 0.02% w/v BSA, pH 6.8) were added to each sample and incubated at room temperature for 15 min in the dark. Resazurin is oxidized by NADH, the product of the reaction between inositol dehydrogenase, NAD ϩ , and myoinositol to the fluorescent compound resorufin. For measurement of the resorufin product, 0.1 M Tris-HCl, pH 9.0 (920 l), was added to each FIGURE 1. Permeability assay, TSF. TSF fractionates liposomes by the exchange of light intraliposomal and heavier extraliposomal solutes through reconstituted hemichannels (see "Experimental Procedures"). It fractionates liposomes into two populations based on channel-mediated permeability to urea (U) and sucrose (S), one population of liposomes with functional channels and another in which the liposomes do not contain functional channels. Connexin channels are incorporated into the membranes of unilamellar liposomes and tracer molecules of interest entrapped in the liposomes. When centrifuged in an iso-osmolar density gradient, movement of gradient solutes (typically urea and sucrose) through open channels causes liposomes to become denser and move to a position deep in the gradient. Liposomes without functional channels remain in the upper part of the gradient. Tracers that are permeable through the channels are lost from the liposomes, and those that are impermeable are retained. The average number of channels per liposome is less than 1, so there is always an upper band without channels to serve as a control of tracer loading and nonspecific leakage. Selective permeability among the tracers was determined by direct comparison of the tracer per liposome retained by the two populations of liposomes.
sample and thoroughly vortexed, and the fluorescence was measured ( ex 565 nm and em 585 nm).
Sample blanks were treated identically to test samples, except the inositol dehydrogenase was omitted. Blank fluorescence was subtracted from the sample fluorescence. Measurement of resorufin fluorescence enabled the amount of myoinositol or dephosphorylated inositol phosphate in the samples to be determined by interpolation from an inositol and tracer standard curve. A separate standard curve was constructed for each tracer compound used. This method was applied identically for each of the IPs tested.
Tracer Analysis, Radioimmunoassay-A second assay was used, specific to (1,4,5)-IP 3 (Biotrak TRK 1000-assay kit; Amersham Biosciences). In brief, (1,4,5)-IP 3 recovered from TSF bands competes with a fixed amount of (1,4,5)- Protein Recovery for SDS-PAGE, Western Blotting, and Gold Staining-Immunopurified connexin preparations or connexin recovered from TSF bands (52) were separated by SDS-PAGE. Western blots on polyvinylidene difluoride membranes were stained with specific monoclonal antibodies to Cx26, Cx32, or the HA(HN) 6 epitope, which were detected with alkaline phosphatase-conjugated secondary antibody and nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate substrate (Pierce). For gold staining, nitrocellulose membranes were blocked (10 mM phosphate-buffered saline, pH 7.4, 0.3% v/v Tween 20) for 30 min at 37°C. Following washing, membranes were stained with Colloidal Gold Total Protein Stain (Bio-Rad). The blots were washed with double-distilled water and air-dried. The Cx26/Cx32 ratio of the reconstituted channels in the lower TSF liposome band, i.e. the population of functional channels, did not differ from that in the upper liposome band, i.e. nonfunctional channels, or the nonreconstituted bead elution fractions.
Data Analysis-The ratio of myoinositol or inositol phosphate to liposome fluorescence (rhodamine-PE) was determined for each sample and corrected for the amount of tracer bound nonspecifically to liposomes. The tracer/rhodamine-PE fluorescence ratio of the upper band reflected the net amount of tracer entrapped per liposome prior to TSF and serves as an internal control for tracer trapping and nonspecific leakage from the liposomes. This was normalized as 100% retention of tracer. For the lower band (liposomes containing functional channels), the tracer/rhodamine-PE fluorescence ratio was normalized to that of the upper band. The data are presented as percentages of liposomes that are permeable to a tracer. Permeability data are expressed as mean Ϯ S.E. Statistical significance was tested by compared means using oneway analysis of variance with p Ͻ 0.05.
Correction for More than One Channel per Liposome-For a given ratio () of functional channels to lipid in the liposomes, a Poisson distribution describes the fraction of the liposomes that have functional channels (34). is estimated from the maximum activity (percentage ( p) of liposomes with n active channels; Equation 1).
As increases, the percentage of liposomes with two or more channels increases. Therefore, the tracer retention/loss in the TSF lower band (the liposomes containing functional channels) may not accurately reflect the tracer permeability per channel (e.g. if there are two channels, both would have to be impermeable to a tracer for it to be retained, potentially resulting in greater permeability per liposome than per channel). By using the Poisson distribution, was used to calculate the distribution of n channels in the liposome population, which was used to compensate for the error introduced by some liposomes containing more than one functional channel. This calculation transforms the fraction of tracer-permeable liposomes in a population into an index of tracer-permeable channels.
Modeling of Isoform Stoichiometry Distributions-Assuming independent association of connexin monomers during hemichannel formation, the frequency of each heteromeric channel stoichiometry formed by two connexins is given by a Bernoulli distribution (Equation 2), where a and b are the incorporation probabilities (i.e. relative abundances) of Cx26 and Cx32, and n is the number of subunits in a hemichannel taken r subunits at a time (20,53). For these calculations, n ϭ 6 (as hemichannels are hexameric) and r was 0 Յ r Յ 6.
Molecular Modeling of Permeants-Molecular models of myoinositol and all inositol phosphates used in this study were made using Spartan 04 (Wavefunction, Irvine, CA). When possible, tracer.pdb files were obtained from a data base of small heteromolecules and built using Spartan 04 if unavailable. Energy-minimized conformations were determined using MMFF94 force field rules (54). Hartree-Fock 6 -31G* ab initio models (55), subject to the MMFF94 geometry constraints, were used for predicting molecular orbital structure (0.002 Å 3 /atomic unit) and chemical behavior, allowing electrostatic potential and local ionization potential to be mapped onto the orbital surfaces.

Populations of Heteromeric Channels with Different Aggregate Ratios of Cx26 to Cx32 Obtained from Heterologous and Native Expression Systems
From HeLa Cells-We recently described an expression strategy for recombinant connexin channels that mimics the natural structural heterogeneity of Cx26 and Cx32 channels (30). Homomeric and heteromeric rCx26 and/or rCx32 hemichannels are expressed in communication-incompetent HeLa cells using bi-directional Tet-On-inducible vectors. For homomeric expression, the channels are tagged at the carboxyl terminus (CT) with a cleavable hemagglutinin epitope preceding a (His-Asn) 6 sequence, a HA(HN) 6 tag. For heteromeric expression, only one connexin isoform is tagged.
Connexin was purified from the cell lines as described under "Experimental Procedures," using a monoclonal antibody directed against the HA epitope of the tag. This purification protocol yields connexin hemichannels (20,21,34,44,45). Western blots (lanes indicated by *) and gold stain of the material purified from Cx26Tag and Cx32Tag HeLa lines show purification of homomeric Cx26 or Cx32 (Fig. 2a, lanes 1* and 2 and lanes 7 and 8*, respectively). Immunopurification from different Cx26Tag/Cx32 cell lines (or Cx26/ Cx32Tag lines; data not shown) yielded both Cx26 and Cx32. Gold stain densitometry showed that purifications from different clonal heteromeric cell lines yield channels with different aggregate isoform ratios (fractional Cx26 contents 0.75, 0.50, 0.33, and 0.25; Fig.  2a, lanes 3-6). Thus, each co-expressing cell line expressed the two connexins in a distinct and defined ratio.
An additional way to obtain Cx26/Cx32 channels with different aggregate isoform ratios utilized the tendency of heteromeric channels with higher Cx26 content to elute sooner/more easily from the anti-Cx32 affinity column (see "Experimental Procedures"). From rat liver membranes, the first two elution fractions yielded fractional contents of 0.66 rCx26 and 0.50 rCx26, respectively, whereas the third elution fraction contained no detectable rCx26 (Fig. 2b, lanes 9 -11). Only elution fractions that contained no detectable Cx26 were considered to contain homomeric rCx32 channels. Similar fractionation of mouse purifications yielded different mCx26/Cx32 ratios for different elution fractions (Fig. 2b, lanes 14 -16). The first elution fraction yielded fractional content of 0.75 mCx26 and the second 0.50 mCx26, whereas the third had no detectable mCx26.
It should be noted that the aggregate Cx26/Cx32 ratio is the ratio of the two isoforms in a given population of channels. Each population consists of channels with a range of isoform stoichiometries, the average of which is the given aggregate ratio. Thus each heteromeric channel population is heterogeneous with regard to the stoichiometries of individual channels.

Inositol Phosphates as Tracer Compounds for Permeability Analysis
The selectivity of inositol phosphates by homomeric and heteromeric connexin channels was explored. IPs are involved in many aspects of cell regulation, and their metabolic products may also play essential roles in cellular function (37). In addition, myoinositol and IPs with one to four phosphate groups include an excellent set of probes with which to examine the roles of size, charge, and charge distribution on the molecular selectivity of connexin channels (Table 1 and Fig. 3).
The ability of the channels to differentiate among highly similar molecules is striking. These data show that heteromeric channels are selectively permeable among inositol phosphates, whereas the corresponding homomeric channels are not. This suggests a unique role for heteromeric channels in biological signaling utilizing IPs.
These data show that there is little difference in the selective permeabilities as a function of the source of the channels (native versus heterologous) or, surprisingly, as a function of the differences in the aggregate Cx26/Cx32 ratio tested. The data are plotted as functions of Cx26/Cx32 ratio in Fig. 5, a (native) and b (HeLa). The differences in permeability over the range of heteromeric ratios tested were not significant.
The permeability of inositol, 1-IP, and (1,4)-IP 2 through purified connexin hemichannels was also assessed using the same techniques. We found that inositol (Fig. 5, a and b, black line), 1-IP, and (1,4)-IP 2 (data not shown) were permeable through all tested hemichannels with no statistically significant difference between them at any fractional Cx26 content ( p Ͻ 0.05; n ϭ 5).

Permeation of (1,4,5)-IP 3 and (1,4,6)-IP 3 through Channels of Different Stoichiometry from the Same Native Source
The data presented thus far from rodent channels compares channel populations with different aggregate Cx26/Cx32 ratios from different species. To determine whether species differences are a factor in the differences in IP selectivity, rat and mouse channel populations with a range of aggregate Cx26/Cx32 ratios were obtained by the selective elution technique described under "Experimental Procedures." The permeabilities to (1,4,5)-IP 3 and (1,4,6)-IP 3 were assessed and compared with those from HeLa cells.

Heteromeric Channel Stoichiometry Does Not Correlate with Molecular Permeation of (1,4,5)-IP 3 or (1,4,6)-IP 3
The absence or weak dependence of the selective permeability on Cx26/Cx32 ratio over the heteromeric range tested suggests that selectivity to IPs is not a simple function of this ratio. To test this explicitly, the relationship between specific molecular permeation and predicted channel isoform stoichiometry was examined. By assuming independent probability of occupancy of each radial channel position, stoichiometry distributions over a range of aggregate isoform ratios were calculated using Bernoulli trials ( Fig. 7a; Equation 2). Curves were then generated reflecting the fraction of channels with isoform stoichiometries less than a given value, as a function of aggregate ratio (Fig. 7b) (20,53). If the content of a specific connexin isoform determined the permeability of a compound, e.g. if n or fewer of a particular connexin in a channel is required for a specific permeability or impermeability, the permeability data as a function of fractional Cx26 content should follow one of these curves.
This analysis does not consider differences in the arrangements of connexin monomers around the pore at each isoform stoichiometry. The inability of this analysis to account for the permeabilities suggests that they may be functions of specific isoform radial arrangements in the channels (59). No matter what the structural mechanism, for such finely tuned molecular discrimination among IPs to occur only through heteromeric channels suggests that recognition sites for these molecules must exist, i.e. are formed by Cx26 and Cx32.

Molecular Modeling of Myoinositol and Inositol Phosphates
The ability of the heteromeric channels to discriminate (1,4,5)-IP 3 from (1,4,6)-IP 3 and (1,3,4)-IP 3 suggests that discrimination is not on the basis of molecular size, charge, or polarity but rather on the basis of specific thermodynamic interaction with a discrete site(s) within different heteromeric hemichannel pores.
To help support this inference, equilibrium geometries and energyminimized conformations for myoinositol and all IPs used were calculated using MMFF94 force field molecular mechanics (54). Electron density models, i.e. size surfaces of the energy-minimized structures (Fig. 3), were generated for these structures using the ab initio Hartree-Fock self-consistent field method (55). For isomers of IP 3 (Fig. 3, d-f ), molecular modeling shows that volume, surface area, and area of polar versus nonpolar surfaces are identical for (1,4,5)-IP 3 , (1,3,4)-IP 3 , and (1,4,6)-IP 3 (Table 1). However, the molecular shape is clearly different when viewing the structure from above the inositol ring and along the C-1 to C-4 axis (i.e. the location of phosphate groups common to all isomers).
Knowledge of the electron density surface allowed graphical representation of predicted three-dimensional chemical characteristics. The electrostatic and local ionization potentials were mapped onto the electron density surface. The electrostatic potential (Fig. 8, a and c) describes regions of the molecule acting as Lewis acids or bases, i.e. areas of negative (red) or positive charge (blue). For the three isomers of IP 3 , negative charge is present at the three phosphate groups, and the relative differences in charge of the common C-1 and C-4 phosphate groups on each of the isomers are minor. Local ionization potential (Fig. 8, b and d) reflects regions from which electrons are most easily lost, i.e. sites that are most susceptible to electrophilic attack, i.e. are likely to become ionized (red), and conversely sites that are more susceptible to nucleophilic attack (blue).
By using this latter graphical representation, we noticed striking differences between and at the C-1 and C-4 phosphates of all isomers; the C-4 phosphate group of (1,4,5)-IP 3 is predicted to be less readily ionized than the C-4 phosphate group of (1,4,6)-IP 3 or (1,3,4)-IP 3 . In the TSF tracer study, (1,4,5)-IP 3 was the most permeable through heteromeric channels. Thus, the distinctly different ionizability of the C-4 phosphate group of (1,4,5)-IP 3 may be involved in the discrimination among the IP 3 isomers.

DISCUSSION
The essential findings of this study are as follows: (a) that connexin channels can discriminate among highly homologous inositol phosphates, most strikingly among inositol triphosphates, and (b) that only heteromeric channels were capable of this discrimination, with the corresponding homomeric channels being nonselective. These findings have implications for biological signaling and for the mechanisms of connexin channel molecular selectivity.
It was believed for many years that connexin channels were so wide that they would be nonselective among ions and molecules small enough to enter the pore. A large body of data now shows that homomeric channels formed by each connexin isoform are selectively permeable among different molecules of similar size, or by the same molecules at different rates (5)(6)(7)(8)(9)(10)(11)(12)(13). Of particular biological importance is the differential permeability of cytoplasmic molecules, including second messengers (6, 20, 21, 25, 60 -62).
Our previous work, using native channels composed of Cx26 and/or Cx32, indicated that heteromeric channels could discriminate between cAMP and cGMP (20,21). The data suggested that channel composition determined permeability to cyclic nucleotides and discrimination among them. Also in the prior work, we observed a trend between impermeability of (1,4,5)-IP 3 and the presence of Cx26 in heteromeric Cx26/Cx32 channels (20).
Because this earlier work suggested that the stoichiometry of Cx26 and Cx32 within heteromeric channels was a determining factor in the selectivity, it was a goal of the present study to explicitly explore the role of isoform stoichiometry. Ideally, such experiments would be carried out with populations of heteromeric channels of single, defined stoichiometries. Unfortunately, such a system is not available.
In this study, populations of native hemichannels with a range of defined aggregate ratios of Cx26/Cx32 were obtained by immunoaffinity purification from livers of different rodent species and from HeLa cells transfected to co-express two connexins. Purifications from rat, mouse, and guinea pig livers contain different proportions of Cx26 as follows: from rat typically ϳ25% rCx26, mouse livers ϳ33% mCx26, and guinea pig livers ϳ80% gpCx26. Channel populations with different aggregate Cx26/Cx32 ratios were also obtained by the judicious use of specific elution fractions from the immunoaffinity matrix, as described under "Experimental Procedures." Homomeric Cx26, homomeric Cx32 channels, and heteromeric Cx26/Cx32 were also obtained from a HeLa expression system using a carboxyl-terminal purification tag. Because the purification tag was on only one of the co-expressed connexins in FIGURE 7. Permeability is not a simple function of the component stoichiometries. a, for a given aggregate Cx26/Cx32 ratio, the fractional component of individual stoichiometries of the channels can be calculated using binomial-based Bernoulli trials (Equation 1) (assuming occupancy of each of the six radial sites in a hemichannel by Cx26 and Cx32 is independent). The leftmost curve represents the fractional amount of homomeric Cx32 channels. Progressing rightward, each curve represents the fraction of channels in which one Cx26 replaces one Cx32. The rightmost curve represents the fractional amount of homomeric Cx26 channels. b, when the probabilities of forming hemichannels with different Cx26/Cx32 ratios are successively summed, a second ensemble of curves result that should account for functional properties (permeability, in this case) if channel stoichiometry is the sole determining factor. As a function of aggregate Cx26 content, the curves show, from left to right, the fraction of channels containing n or fewer Cx26 monomers. If permeability is a simple function of Cx26 to Cx32 stoichiometry, the data as a function of aggregate Cx26/Cx32 content should follow one of these curves. A vertical abscissa (the aggregate ratio Cx26/Cx32 in a given hemichannel preparation) intersecting the curves indicates the percentage amounts of the respective homomeric or heteromeric channel stoichiometries in the aggregate channel population. c-f, superimposition of the permeability data from Figs. 5 and 6 onto these curves shows that, with the assumption of independent distribution of monomers, the permeability to IPs could not be accounted for by simple stoichiometry distributions. However, the graphs do make the point that (1,4,5)-IP 3 was permeable through fewer of the heteromeric hemichannels as the aggregate ratio of Cx26/Cx32 approached 1:1 and permeable through more of the channels when either isoform predominated. (1,4,6)-IP 3 was permeable through fewer of the heteromeric hemichannels than (1,4,5)-IP 3 at all aggregate isoform ratios. FIGURE 8. Predicted chemical properties of inositol triphosphates. Hartree-Fock 6 -31G* ab initio models, subject to MMFF94 equilibrium geometry constraints, were used for predicting molecular orbital structure (0.002 Å 3 /atomic units; see Fig. 3); electrostatic potential and local ionization potential were mapped onto the orbital surfaces by Spartan 04. The electrostatic potential (a and c) describes regions of the molecule that carry negative (red) or positive charge (blue). Local ionization potential (b and d ) reflects regions sites are most susceptible to electrophilic attack, i.e. are likely to become ionized (red) and, conversely, that are more susceptible to nucleophilic attack (blue). Tracers shown, from top to bottom, are (1,4,5)-IP 3 , (1,4,6)-IP 3 , and (1,3,4)-IP 3 . a and b, IPs are represented with the same orientations and scale as in Fig. 3, as observed from above the inositol ring and en face. c and d, structures are viewed lengthways along C-1 to C-4 of the inositol ring. c, C-4 is in the foreground; d, C-1 is in the foreground. The molecular orbital meshwork structures (0.002Å 3 /atomic unit) for these viewpoints are also shown. heteromeric channels, we were able to purify channels by their Cx26 or Cx32 content, biasing the purification for one or the other connexin. In addition, cell lines were used that expressed the two connexins in different ratios. By use of these methods, we were able to compare the permeability properties of channel populations over the same range of aggregate stoichiometries from each species, from different species, and from a heterologous expression system.
The data show that homomeric Cx32, homomeric Cx26, and heteromeric Cx26/Cx32 channels, irrespective of Cx26/Cx32 aggregate ratio, were permeable to myoinositol, (1)-IP, and (1,4)-IP 2 . Selectivity was observed only when more than two phosphate groups were present on the myoinositol ring and only for the heteromeric channels. In each of the populations of heteromeric channels, the fraction of heteromeric channels permeable to (1,4,6)-IP 3 , (1,3,4)-IP 3 , or (1,3,4,5)-IP 3 was far less than the fraction permeable to (1,4,5)-IP 3 . In the case of (1,4,5)-IP 3 , the channels were most selective when the aggregate Cx26 to Cx32 was close to 1:1. This was not true for (1,4,6)-IP 3 , which had approximately the same permeability through all heteromeric Cx26/Cx32 channels, regardless of stoichiometry. The permeabilities to the IPs were not significantly different for channels purified from the different rodent species or purified from HeLa cells with the same aggregate stoichiometries.
The selectivities we report here could not have been inferred or anticipated from the studies of connexin channel permeability using the many fluorescent tracers that have been used by our group and others (7,8,10,57,60,63). The selectivities were revealed using the biological signaling molecules themselves as probes of the connexin pore. It seems that connexin channels have been finely tuned to select among the cytoplasmic molecules to which they are exposed in situ, and that selectivities among nonbiological tracers are secondary consequences.
The ability of heteromeric Cx26/Cx32 channels to differentiate among different inositol triphosphates ((1,4,5)-IP 3 , (1,4,6)-IP 3 , and (1,3,4)-IP 3 ) was surprising. These isomers have the same molecular weight and charge; the difference is the location of the phosphate groups on the myoinositol ring, which affects shape and charge distribution, and the chemical reactivity of the phosphate groups. This finding emphasizes the importance of the spatial configuration of phosphate groups on the biological selectivity of permeation through connexin channels.
The high degree of selectivity among very similar permeants suggests there are specific sites where permeants interact with the pores. The data are most consistent with a mechanism in which, at least for Cx26/ Cx32 channels and this class of second messengers, a molecule must interact appropriately with a highly selective but likely low affinity binding site in the pore in order to permeate, and that molecules that do not cannot permeate. This is analogous to the mechanism for selectivity for K ϩ over Na ϩ in K ϩ channels, only for signaling molecules instead of monovalent atomic cations.
The requirement for heteromericity suggests that such a site involves structural contributions from more than one monomer and from both Cx26 and Cx32 (59). Examination of known inositol phosphate-binding sites did not reveal amino acid sequence similarity with Cx26 or Cx32 and specifically no correspondence with conserved residues in the (1,4,5)-IP 3 receptor-binding core or the IP 3 -coordinating residues (64). This is not surprising given the requirement for heteromericity and presumed low affinity. In this context it is intriguing to note that an engineered ␣-hemolysin pore, which does not contain a canonical IP 3binding site, was able to bind IP 3 at high affinity, other inositol phosphates with lower affinity, and cAMP not at all (65).
The most unexpected finding was that the homomeric Cx26 channels were also permeable to all the IPs. This is inconsistent with the trend toward more restrictive permeability with increased Cx26 content that we had noted in previous work (20,21). One possibility is that Cx26/ Cx32 channels, because of their heteromeric nature, have an irregular or narrowed lumen relative to both types of homomeric channels, perhaps because of lack of axial registration of the pore-lining segments of the two isoforms, and these features enable the observed selectivity.
The results make the important point that the earlier work with cyclic nucleotides was not a singular result, and native channels consisting of these two isoforms can be selectively permeable to two classes of second messengers and can distinguish among the members of each class. From the present data we cannot draw a detailed relation between aggregate isoform stoichiometry and the selectivities, but the data strongly indicate that dynamic control of isoform composition of connexin channels in cells (66 -68) can dramatically modulate intercellular signaling mediated by the two major classes of second messengers. Specifically, for Cx26 and Cx32, which are often expressed in the same cells and in the same channels (20,21,69,70), the data suggest that cells could decrease selectivity among IPs by down-regulating either connexin and can enhance selectivity by making both proteins in equal amounts.

CONCLUSION
Our findings suggest a unique functional role of heteromeric channels, with implications for connexin channel cell biology. The emerging theme is that heteromeric connexin channels can have a remarkable range of selectivities among cytoplasmic molecules that is not predicted from conventional dye permeability studies. The size and charge of a permeant alone do not alone determine permselectivity because the heteromeric pore can have a significant selectivity among biological permeants with the same size and overall charge. Our results provide a ready explanation as to why more than one connexin is often expressed in the same cells and why they form heteromeric channels. Heteromeric connexin hemichannels can be restrictive and selective, at least regarding IP 3 and cyclic nucleotide permeability, whereas the corresponding homomeric channels are not. The specific selectivities to IPs depend on connexin channel composition and likely isoform arrangement within channels. The next major challenge lies in determining the structural basis of the striking selectivity of different connexin channels.